You Are Who You Follow

Mathew Ingram reports on the continuing efforts of various web services to decipher influence:

Mathew Ingram, The Race to Build a PageRank for the Social Web Continues

[…]

Both PeerIndex and Klout rank users based on data that comes from their Twitter, Facebook and LinkedIn accounts, although the two sites describe their rankings somewhat differently. Klout talks about overall “reach” and “amplification,” both of which are determined by looking at a user’s activity and how much impact it has on their social graph — whether their tweets are re-tweeted by others with influence, for example. PeerIndex says that it looks at a user’s activity in Twitter, Facebook and LinkedIn and then comes up with an authority rank for their expertise in eight topic areas, which it uses to create an influence “footprint” for each user.

[…]

As Klout and PeerIndex add more sources of reputation or influence data such as Quora to their rankings, the web moves closer to having a kind of Google PageRank for social activity, with all that implies. The big problem, as with Google search, is how to exclude the social equivalent of black-hat SEO and link spam, and how to determine what it is real influence and what is simply Justin Bieber-style popularity.

You are who you follow.

I continue to believe that these tools are motivated by a skewed model of what is going on online. It’s a sort of neo-classical economics viewpoint, where are individuals are considered as interchangeable, like replacement parts on an assembly line, or atoms banging into each other in a tea kettle of boiling water. And worse, they are considered in isolation, not as connected to others in a deep way.

We are social beings, and our social value is an emergent property of the network we are situated in, not a personality trait. We need to think about this issue in terms of position in the social network — who we follow and are followed by — not purely statistically. 

Stowe Boyd, It’s Betweenness That Matters, Not Your Eigenvalue

Betweenness is a measure of how short are the chains that connects a person to the totality of the network. Like PageRank, betweenness is recursive: the people with the highest betweenness are likely to be connected to other people with high betweenness.

This means people are influential because they are connected to many influential people. But influence doesn’t seem directly linked to how many people you are connected to. It’s a function of being connected to others who have short chains to many other people with high betweenness. Or, looked at differently, betweenness is a measure of how many social circles, or social scenes, a person is connected to.

So, it’s not who you know, it’s where you know. It’s where you are situated in the network, and not just in the limited sense of how many immediate contacts you have.

The subtle, dark-matter mystery of social networks is that influence is oblique, and not easily determined by the sorts of tools we have today.

It is not your follower count, or who you follow, per se. But, instead, do you have short paths into other social scenes, both incoming and outgoing? That is the deep structure of being truly connected: bridging over different social scenes, acting as a conduit, a vector, a filter and amplifier for ideas good and bad, the best insights, and deadly viruses.

I maintain that the best predictor of an individual’s social value and the likelihood that their social activities will make an impact on others is who they are connected to, and in particular, who they follow. Following people from a diverse range of social scenes — different parts of the global social network — increases the likelihood that you will encounter new insights, new thoughts. And following more people who are connected themselves to diverse social scenes increases that likelihood. And that increases your social value to others.

You are who you follow.

Attempts to quantify our influence — our social value — without taking this into account are doomed to be a second order reflection of what is actually happening between us. It’s like trying to figure out why someone threw a rock into a stream by calculating the rock’s weight, trajectory, and the spread of the ripples. All very interesting, but it doesn’t get to why.

3 March 2011 47 Permalink

Content, Context, Conduit: It’s Not Who You Know, But Where You Know

The other night I was a participant on TummelVision, and one topic that came up was influence. I digressed and starting talking about betweenness, stating that being connected to people in very different ‘scenes’ is much more important than popularity in influencing people:

Stowe Boyd,  It’s Betweenness That Matters, Not Your Eigenvalue: The Dark Matter Of Influence

The most connected people in a social network — those with the highest number of incoming and outgoing connections — have high eigenvalues. These eigenvalues can be calculated — like Google’s PageRank algorithm — by weighting the value of each connection based on the eigenvalue of the originator.

But this research [ARVIX blog, Best Connected Individuals Are Not the Most Influential Spreaders in Social Networks] suggests that a different way to measure the centrality might be more useful in determining how much throw weight a person actually has. Betweenness is a measure of how short are the chains that connects a person to the totality of the network. Like PageRank, betweenness is recursive: the people with the highest betweenness are likely to be connected to other people with high betweenness.

it’s not who you know: it’s where you know. It’s where you are situated in the network, and not just in the limited sense of how many immediate contacts you have.

This means people are influential because they are connected to many influential people. But influence doesn’t seem directly linked to how many people you are connected to. It’s a function of being connected to others who have short chains to many other people with high betweenness. Or, looked at differently, betweenness is a measure of how many social circles, or social scenes, a person is connected to.

So, it’s not who you know: it’s where you know. It’s where you are situated in the network, and not just in the limited sense of how many immediate contacts you have

I think this also is closely related to the notion of curation, which is being discussed a great deal these days, including the recent Business Insider Ignition conference:

Jared Keller, Who Do You Want Telling You What to Read?

I listened to John Borthwick of startup incubator Betaworks, Garrett Camp of discovery engine StumbleUpon, Patrick Keane of Associated Content, and Mark Josephson of the hyperlocal Oustide.in discuss the merits of algorithmic and crowdsourced modes of navigating the news. They’re all like me: they primarily discover content through a carefully curated Twitter feed, an RSS reader, or some other social news service.

So which is better? It’s usually a mix of the three. “Technology, math and algorithms are being used to refine and understand how people filter what they are looking at and how they read,” Borthwick said. “But mainly people read the voice of other people. There are new tools for getting there, so content production is being pushed into the pale, but most of these tools when they are used well are used to surface and filter, not compose.”

“In social media, everyone should be a content creator and curator,” Camp added. “StumbleUpon is trying to blend both worlds by asking for human input on thumbing stories up and down.”

In this sense, algorithms aren’t replacing editors or individual voices, but are used out of necessity: as the cost of creating content continues to drop, the sheer amount of content available to consumers has exploded. Algorithms are just there to lead the way. And sometimes those algorithms help us find content that, while not produced professonally, has an incredible amount of value.

So it’s not the content, which is the readage. Nor the context, as delivered by better metadata. And it’s not even the act of picking what is best, and passing it along, which is what most people mean by curation. What is really core are the actions taken to increase betweenness, by adding conduits to others who are well-connected and dropping ones that are less well-connected.

I am not sure how ‘carefully curated’ these folks’ twitter streams are, actually, but leave that aside. It is obvious that there is a very fast movement away from search and RSS, and toward Twitter as a source of readage for the world of media and mediaheads.

Instead of considering what we are doing as filtering, or curating, we should think about tinkering with our connections as a way to position ourselves in the network to maximize certain characteristics for the nodes we occupy.

Take betweenness, for example. I try to follow people who are connected to very well-connected people, where ‘well-connected’ does not mean popular, per se, but instead means connected to many people in different walks of life, different countries, different jobs.

The outcome of this tinkering with my connections is that I increase my betweenness: I shorten the number of links than connect me to the entire network, and the world.

Looking at it another way, I am also increasing my utility to those that follow me by increasing my betweenness: they are more likely to find unique insights or innovative ideas by following me, since I bridge many social scenes.

It’s a virtuous cycle: by adjusting my position in the network — by following and unfollowing — I improve the diversity and quality of readage I see, and by passing along the best of what comes along, my followers are better off. My actions improve their respective positions in the network too. And those of their followers, and so on. I am actually improving the entire network, by better positioning myself.

So it’s not the content, which is the readage. Nor the context, as delivered by better metadata. And it’s not even the act of picking what is best, and passing it along, which is what most people mean by curation. What is really core are the actions taken to increase betweenness, by adding conduits to others who are well-connected and dropping ones that are less well-connected.

And, no, there aren’t any algorithmic solutions out there for this. It would be handy if Twitter or Klout would offer up a betweenness value for a given user, or offer up some recommended connections based on the principle of positioning myself better in the network, not just because of topical relevance, or popularity. But they don’t.

Could you get on that guys?

Source: The Atlantic

4 December 2010 15 Permalink

I am featured on paper.li, or, er, my upstream is

Paper.li is self-described in this way: 

paper.li organizes links shared on Twitter into an easy to read newspaper-style format. Newspapers can be created for any Twitter user, list or #tag.

Jamie Burke pinged me today, pointing out that my paper.li ‘daily journal’ is at the top of the service’s featured page.

I don’t know how the featured list is composed, but It’s kind of an odd feeling to be on the list since paper.li is simply ‘journalizing’ things that people I follow are pointing to. It’s not my own writing abilities, here, or reasoning. It’s my curatorial ability: I am following people whose links turn out to be interesting to others. So my upstream is being featured on paper.li, really.

Wait a minute! That’s the whole thesis of this piece I wrote not too long ago, called It’s Betweenness That Matters, Not Your Eigenvalue: The Dark Matter Of Influence. I make the case that the people that are likely to be the most influential aren’t the obviously most popular ones:

The most connected people in a social network — those with the highest number of incoming and outgoing connections — have high eigenvalues. These eigenvalues can be calculated — like Google’s PageRank algorithm — by weighting the value of each connection based on the eigenvalue of the originator.

But this research [Maksim Kitsak, et al] suggests that a different way to measure the centrality might be more useful in determining how much throw weight a person actually has. Betweenness is a measure of how short are the chains that connects a person to the totality of the network. Like PageRank, betweenness is recursive: the people with the highest betweenness are likely to be connected to other people with high betweenness.

This means people are influential because they are connected to many influential people. But influence doesn’t seem directly linked to how many people you are connected to. It’s a function of being connected to others who have short chains to many other people with high betweenness. Or, looked at differently, betweenness is a measure of how many social circles, or social scenes, a person is connected to.

So, it’s not who you know it’s where you know. It’s where you are situated in the network, and not just in the limited sense of how many immediate contacts you have.

My hunch is that my upstream is interesting to people because I am following people who themselves are connected to very connected people, people from a wide range of social scenes.

16 August 2010 Permalink

Jeff Jarvis on The Hunt For The Elusive Influencer

Jeff Jarvis is right when he makes the point that those with the most followers may not be the most influential; but he misses the fact that some people might still be more influential than others:

Jeff Jarvis, The Hunt For The Elusive Influencer

[…] trying to find the big influencer with big audience is really just old mass marketing in a cheap dress. Old mass marketing (go with the largest numbers … and breasts) isn’t economical; neither, it turns out, is marketing to just one or a few powerful people — the mythical influencer. That brings us to a new hybrid to mass marketing, which is what I think Watts is suggesting: Target many people who at least have some friends who’ll hear them. (Disclosure: This was a key insight in the development of the company 33Across that made me invest in it.)

Or to put this question in the current argot: Is there more influence in the tail than in the head? If you talk to 100k people who talk to 10 people each, do you get more bang than talking to one person who has 1m followers? (Watts did also say that a combination of mass and tail marketing is effective.)

Just because the most popular people are not the most influential does not mean that no one is influential. Jarvis seems to fall back to a position that there are no influencers:

So the message spreads not because of who spoke it but because the message is worth spreading. What makes us spread it? First, again, we spread it if it resonates and it is relevance; it has value to us and we think it will have value to others. Second, trust or authority is a factor. If I see Clay Shirky or Jay Rosen or Kevin Marks tell me to click on a link I’m more likely to do so because I respect them and trust their judgment and I’ve found in the past that clicking on their links tends to be worth the effort. They give me ROC (return on click). But if I followed Miss Kardashian (I don’t) and she told me to click on a link, I’d be less likely to, both because I don’t put her in the same intellectual corral as my other friends and have no relationship with her and because I have seen that clicking on her links gives me lousy ROC. Is trust or authority or experience influence? In a small circle of actual friends, I don’t think so. And in any case, having only a small circle of friends isn’t the one-stop-shopping influence marketers are seeking.

So abandon the hunt, marketers. You’re not going to bag the influencer. She doesn’t exist (well, one did but she quit her TV show).

The flaw in his argument is that popularity is not the only way to weight nodes in a social network. Jarvis mentions authority, but doesn’t go very far with his analysis. However, I think authority is a red herring, too. It is looking at the transmission of messages int he context of an individual’s value judgments, as if we decide what to be influenced by, day by day.

But influence is actually a sort of dark matter: a force that surrounds us without us really being aware of it, like gravity.

In recent posts, I have explored these ideas at some length (see It’s Betweenness That Matters, Not Your Eigenvalue: The Dark Matter Of Influence and Social Scenes: The Invisible Calculus Of Culture), so I will simply reprise some of the recent research about social influence and what it means to us, as individuals.

The number of followers a person has is an indicator of a sort of connectedness in a social network, but it is not a good predictor of influence. Even when you weight the value of each link based on the rank of each person connecting to someone, like Kim Kardasian, it doesn’t line up with influencing others. That weighted measure is — to use technical terms for a second — called the eigenvalue, and while it is a measure of ‘centrality’ — a degree of connectedness in the network — centrality isn’t the measure of centrality that best aligns with influence.

Another form of centrality is to look at where an individual sits in the network relative to subnetworks. For example, a person who has solid connections in the tech community and a number of deep connections in the art world is likely to act as a bridge between those communities, and carry new ideas from art to tech and vice versa. This is called ‘betweenness’. To the degree that people that traverse different social scenes are rare, and if these communities benefit from this cross-pollination, then such bridgers will have an inordinate influence of both communities. But it is insufficient to simply measure the number of social scenes that an individual touches, just as it is inadequate to simply count links to a page to determine its page rank: you have to weight the links by the ranking of the pages that link. That’s the core of Google’s PageRank algorithm. However, in the case of this sort of bridging across social scenes, the individual’s betweenness has to be calculated based on the sum of the betweenness of all those that she connects to. In this way, one person’s betweenness is a function of the betweenness of all of her connections.

This seems intuitive: people who have many connections into diverse social scenes will act as the conduit for ideas to spread. And if I am connected to many others who are likewise connected to diverse social scenes, then I am even more likely to spread ideas: I am a better idea vector to the extent that I have more of these kinds of connections: more betweenness.

The conclusion here is that betweenness is a good predictor of influence, because influence is strongly linked with exposing people to new ideas, trends, or culture, while eigenvalues are not. The most popular person in a social scene may not be the one speanding a lot of time in other social scenes; on the contrary.

And the final piece of the puzzle is that we are all embedded in social scenes that are larger than we know. For example, I am connected to hundreds of friends, who are influenced by tens of thousands of their friends, some of which I may know, and more of which I could encounter. However, the friends of my friends are influenced by the third closure, the social scene of millions of people, a scene so large I cannot possibly know all those involved.

Recent research has shown that it is this social scene scale that influences our weight, our health, our smoking habits. This dark matter — the third closure — influences us like an atmosphere: we don’t notice it, but it is filling our lungs, and pressing against our skin. We meet some friends who mention a new sort of club music, and a few hours later you hear some on your favorite radio station. The next day, at a friends’ house, she’s playing the same band on her stereo, and that night you hear it again at your favorite bar.  That’s because in your corner of the galaxy, some number of people with high betweenness, floating around in the third closure, dragged this new music into the tech scene from the art scene, and turned a bunch of people onto it, and a year later it’s on the pop radio channels.

So, people need to bone up a little on network research to get the differences between different sorts of centrality, and to unthread popularity from influence, mathematically and anthropologically. Just because popularity isn’t a good predictor of influence doesn’t mean nothing is.

Betweenness and the dark matter of the third closure are the keys to understanding — and potentially directing — how influence flows though social networks. There’s a lot to research yet, but these will likely be the starting points of influence science going forward.

5 April 2010 23 Permalink

It’s Betweenness That Matters, Not Your Eigenvalue: The Dark Matter Of Influence

Let me explain, before you think I have been gargling alphabet soup.

Recent research suggests that the most important people in social networks, relative to actually transmitting ideas, viruses, or moods, might not be the folks with the most followers, but instead might be people that are connected to a large number of individuals through shorter paths than others have.

- ARVIX blog, Best Connected Individuals Are Not the Most Influential Spreaders in Social Networks

The study of social networks has thrown up more than a few surprises over the years. It’s easy to imagine that because the links that form between various individuals in a society are not governed by any overarching rules, they must have a random structure. So the discovery in the 1980s that social networks are very different came as something of a surprise. In a social network, most nodes are not linked to each other but can easily be reached by a small number of steps. This is the so-called small worlds network.

Today, there’s another surprise in store for network connoisseurs courtesy of Maksim Kitsak at Boston University and various buddies. One of the important observations from these networks is that certain individuals are much better connected than others. These so-called hubs ought to play a correspondingly greater role in the way information and viruses spread through society.

In fact, no small effort has gone into identifying these individuals and exploiting them to either spread information more effectively or prevent them from spreading disease.

The importance of hubs may have been overstated, say Kitsak and pals. “In contrast to common belief, the most influential spreaders in a social network do not correspond to the best connected people or to the most central people,” they say.

At first glance this seems somewhat counter-intuitive but on reflection it makes perfect sense. Kitsak and co point out that there are various scenarios in which well connected hubs have little influence over the spread of information. “For example, if a hub exists at the end of a branch at the periphery of a network, it will have a minimal impact in the spreading process through the core of the network.”

By contrast, “a less connected person who is strategically placed in the core of the network will have a significant effect that leads to dissemination through a large fraction of the population.”

The question then is how to find these influential individuals. Kitsak and co say that the way to do this is to study a quantity called the network’s “k-shell decomposition”. That sounds complicated but it isn’t: a k-shell is simply a network pruned down to the nodes with more than k neighbours. Individuals in the highest k-shells are the most influential spreaders.

(via @karllong)

In network theory, these two cases are both example of centrality: ways of assigning values to individual nodes in a network based on how each node relates to the others.

The most connected people in a social network — those with the highest number of incoming and outgoing connections — have high eigenvalues. These eigenvalues can be calculated — like Google’s PageRank algorithm — by weighting the value of each connection based on the eigenvalue of the originator.

It’s not who you know, it’s where you know.

But this research suggests that a different way to measure the centrality might be more useful in determining how much throw weight a person actually has. Betweenness is a measure of how short are the chains that connects a person to the totality of the network. Like PageRank, betweenness is recursive: the people with the highest betweenness are likely to be connected to other people with high betweenness.

This means people are influential because they are connected to many influential people. But influence doesn’t seem directly linked to how many people you are connected to. It’s a function of being connected to others who have short chains to many other people with high betweenness. Or, looked at differently, betweenness is a measure of how many social circles, or social scenes, a person is connected to.

So, it’s not who you know, it’s where you know. It’s where you are situated in the network, and not just in the limited sense of how many immediate contacts you have.

The subtle, dark-matter mystery of social networks is that influence is oblique, and not easily determined by the sorts of tools we have today.

It is not your follower count, or who you follow, per se. But, instead, do you have short paths into other social scenes, both incoming and outgoing? That is the deep structure of being truly connected: bridging over different social scenes, acting as a conduit, a vector, a filter and amplifier for ideas good and bad, the best insights, and deadly viruses.

related articles

• Location determines social network influence, CCNY-led team finds (scienceblog.com)
• Proximity to Core Determines Social Network Influence (chrisabraham.com)

3 February 2010 36 Permalink